Should You Replace Customer Service With AI in 2026?

Uber cut 10% of its customer service team on July 22 and said the quiet part out loud: AI did it. Commonwealth Bank of Australia has cut nearly 800 roles this year with AI cited as a driver, its offshore support contractors among them. Klar

Uber cut 10% of its customer service team on July 22 and said the quiet part out loud: AI did it. Commonwealth Bank of Australia has cut nearly 800 roles this year with AI cited as a driver, its offshore support contractors among them. Klarna ran this exact play in 2024, and by May 2025 it was quietly hiring humans back.

Same technology. Three different companies. Two very different outcomes.

If you're a small business owner wondering whether to replace customer service with AI, 2026 just handed you a live experiment at a scale you'll never have to run yourself. Here's what it actually says.

Three companies just proved AI can replace a support team

Uber's cut hit its community operations team specifically, the group handling rider and driver disputes. Megha Yethadka, the VP running that department, put it plainly: "We cannot scale frontier technology on top of fragmented processes." Translation: the AI worked well enough that the old headcount stopped making sense, according to Bloomberg's reporting.

Commonwealth Bank of Australia has been running the same play all year. Its Hey CommBank chat platform was resolving close to nine in ten customer conversations without any agent involvement by May 2026. The Finance Sector Union has tracked roughly 800 total role cuts across 2026 tied to that rollout, including 276 redundancies in July alone. CBA disputes that framing and says it has hired more than 140 contact-centre staff in Australia this year, but the offshore support contractors whose chat work the AI absorbed are gone either way.

Zoom out and this isn't a fringe bet anymore. Salesforce's State of Service report, a survey of more than 6,500 service professionals published in November 2025, put AI's share of resolved cases at 30% today and projected 50% by 2027. Forrester goes further: its analysts project that by 2030, AI will eliminate 49% of current customer service jobs, weighted toward high-volume operations where the exceptions are rarer and more repeatable.

Klarna already ran this experiment, and reversed it

Klarna got there first, and it's the cautionary half of this story. By early 2024 the company said its AI assistant was doing the work of 700 full-time agents, and it hired zero new human support staff that year.

As cost unfortunately seems to have been a too predominant evaluation factor when organizing this, what you end up having is lower quality.

— Sebastian Siemiatkowski, Klarna CEO, to Fortune

In May 2025, Siemiatkowski reversed course. Customers were repeat-contacting for the same unresolved issue, satisfaction scores had slipped, and the savings on paper weren't showing up in the outcomes that mattered. Klarna is now hiring humans again and pitching real conversations as a premium feature, not the thing it automated away.

The lesson isn't "AI support fails." Uber and CBA are proof it doesn't, at scale, on the right kind of ticket. The lesson is that going all-in on either extreme, all-human or all-AI, is where the money actually gets lost.

The math depends on ticket volume, not company size

This is where most small business owners misread the 2026 headlines. Uber and CBA aren't cutting headcount because AI got smart. They're cutting it because they have the ticket volume to make automation statistically safe: thousands of near-identical questions a day, where a bad AI answer is one blip in a system built to catch it.

A ten-person service business doesn't have that volume, and that changes the calculation completely. If you're fielding 40 support conversations a week and half of them are genuinely different from each other, an AI-only system doesn't have enough repetition to learn the pattern before it costs you a customer.

That's not a reason to skip AI support. It's a reason to build the escalation layer first, not last, which is exactly what we cover in the grounding setup that keeps AI support from inventing answers.

The hybrid model is the only one still standing after a year

Every "go 100% AI" brief we've gotten from a client has turned into a hybrid build within the first working session, once we walk through what a single wrongly-handled refund or missed contract clause actually costs against the ticket fees saved. Klarna learned that lesson at enterprise scale and in public. Most small businesses can learn it in a planning call instead.

The version that survives contact with real customers looks the same whether you're Klarna-sized or five people: AI resolves the repeatable stuff on its own, drafts a response for anything borderline, and never sends the final word on money, legal exposure, safety, or an already-angry customer without a human seeing it first. That's not a compromise position. It's the only version of "replace customer service with AI" that has actually held up past the first two quarters, for a company of any size.

Deciding what qualifies as "repeatable enough to automate" is its own project, and it's usually the first automation call we get asked to make. We've written up how we help a business decide what to automate first, and the same triage logic that works for lead routing works here: narrow the AI's job, name its exceptions, and put a human on anything that isn't one of the cases you already trust it to close.

None of this requires the volume Uber or Commonwealth Bank has to be worth building. It requires deciding, honestly, which of your support conversations are actually repeatable before you let AI answer them alone. That's the same discipline behind routing and qualifying leads before a human touches them, and it's the difference between the businesses quietly saving money on support this year and the ones quietly writing an apology email to customers next year.